Ansari Saleh Ahmar
This study examines whether hybrid trend-decomposition forecasting outperforms classical autoregressive methods for inflation prediction in emerging economies experiencing supply-shock-induced regime changes. Monthly year-on-year inflation data published by Bank Indonesia — spanning September 2021 through October 2024 (n = 50 observations) — provide the empirical basis. The 40/10 in-sample/out-of-sample split is designed to ensure that the training window captures the full trajectory of Indonesia's 2022 commodity supply shock (Russia-Ukraine invasion, August 2022 peak: 7.71%) alongside its subsequent policy-driven disinflation, while the evaluation window coincides with the structurally distinct recovery phase (January-October 2024, 2.13-3.22% range). BetaSutte — a parsimonious hybrid model that decomposes inflation into a trend component and a remainder, then applies asymmetrically weighted exponential smoothing to each — achieves an out-of-sample RMSE of 0.352% and MAPE of 0.114%, compared with ARIMA(1,1,1)'s 0.538% and 0.170%, representing reductions of 34.6% and 32.9%, respectively. Crucially, BetaSutte's in-sample RMSE (4.009%) substantially exceeds ARIMA's (2.318%), yet the rankings reverse decisively out-of-sample — a manifestation of the bias-variance trade-off in time-series model selection. This reversal is explained through three mechanisms: deliberate trend-signal extraction over noise fitting, implicit handling of structural breaks via the slope of the least-squares trend line, and avoidance of the differencing operator's destruction of long-term directional information. The Diebold-Mariano test confirms BetaSutte's superiority at p < 0.10. To the best of the author's knowledge, this is the first application of BetaSutte to central bank inflation data in an emerging market setting. Policy implications suggest that central banks in commodity-dependent economies should prioritise out-of-sample accuracy criteria when selecting forecasting tools, particularly when supply-shock episodes are anticipated. © 2025, PT Mattawang Mediatama Solution. All rights reserved.
Department of Statistics, Universitas Negeri Makassar, Makassar, Indonesia
